Daily briefing

Papers fetched on 2026-07-17

Executive Signal

2026-07-17 is led by SEED: Self-Evolving On-Policy Distillation for Agentic Reinforcement Learning, Spectral Rewiring for Exploration, Purification, and Model Merging, and VideoChat3: Fully Open Video MLLM for Efficient and Generalist Video Understanding, with the strongest papers skewing toward production-minded advances that pair novelty with implementation value.

Top Papers

99/100Read

SEED: Self-Evolving On-Policy Distillation for Agentic Reinforcement Learning

Published 2026-07-16 · Fetched 2026-07-17

Innovation Summary

SEED: Self-Evolving On-Policy Distillation for Agentic Reinforcement Learning: We propose SEED (SElf-Evolving On-Policy Distillation), a self-evolving framework that converts completed on-policy trajectories into training-time hindsight skills and distills their behavioral effect back into the.

Executive Summary

SEED: Self-Evolving On-Policy Distillation for Agentic Reinforcement Learning: We propose SEED (SElf-Evolving On-Policy Distillation), a self-evolving framework that converts completed on-policy trajectories into training-time hindsight skills and distills their behavioral effect back into the. Why it matters: Overall signal 99/100 driven by novelty 100 and practical impact 100. It maps to cross-cutting AI systems work even without explicit category metadata. Community signal includes 55 upvote(s) and 1 comment(s), which helps separate durable interest from title-only curiosity. Implementation angle: Implementation potential scores 99/100; prioritize adaptation paths for internal agent, evaluation, or platform workflows. No linked repository is present, so expect more translation work before the ideas are production-ready. Technical depth scores 100/100, so a quick skim should focus on architecture, data, and evaluation sections before full adoption work. Caveat: No linked implementation is available yet, which raises integration cost and lowers reproducibility confidence.

Why It Matters

  • Overall signal 99/100 driven by novelty 100 and practical impact 100.
  • It maps to cross-cutting AI systems work even without explicit category metadata.
  • Community signal includes 55 upvote(s) and 1 comment(s), which helps separate durable interest from title-only curiosity.

Implementation Angle

  • Implementation potential scores 99/100; prioritize adaptation paths for internal agent, evaluation, or platform workflows.
  • No linked repository is present, so expect more translation work before the ideas are production-ready.
  • Technical depth scores 100/100, so a quick skim should focus on architecture, data, and evaluation sections before full adoption work.

Caveat

No linked implementation is available yet, which raises integration cost and lowers reproducibility confidence.

Estimated Reading Priority

High - 99/100 signal; read before acting on adjacent agent, evaluation, inference, or ML systems work.

Links

N/AJSON
99/100Read

Spectral Rewiring for Exploration, Purification, and Model Merging

Published 2026-07-03 · Fetched 2026-07-17

Innovation Summary

Spectral Rewiring for Exploration, Purification, and Model Merging: We show that the reasoning-effective component of these updates is largely concentrated in the base model's spectral space, motivating Subspace-Aligned Rewiring (SAR), a post-hoc editing method.

Executive Summary

Spectral Rewiring for Exploration, Purification, and Model Merging: We show that the reasoning-effective component of these updates is largely concentrated in the base model's spectral space, motivating Subspace-Aligned Rewiring (SAR), a post-hoc editing method. Why it matters: Overall signal 99/100 driven by novelty 100 and practical impact 100. It maps to cross-cutting AI systems work even without explicit category metadata. Community signal includes 16 upvote(s) and 2 comment(s), which helps separate durable interest from title-only curiosity. Implementation angle: Implementation potential scores 100/100; prioritize adaptation paths for internal agent, evaluation, or platform workflows. No linked repository is present, so expect more translation work before the ideas are production-ready. Technical depth scores 100/100, so a quick skim should focus on architecture, data, and evaluation sections before full adoption work. Caveat: Evidence appears benchmark-centric, so verify transfer to production workloads before acting on the claims.

Why It Matters

  • Overall signal 99/100 driven by novelty 100 and practical impact 100.
  • It maps to cross-cutting AI systems work even without explicit category metadata.
  • Community signal includes 16 upvote(s) and 2 comment(s), which helps separate durable interest from title-only curiosity.

Implementation Angle

  • Implementation potential scores 100/100; prioritize adaptation paths for internal agent, evaluation, or platform workflows.
  • No linked repository is present, so expect more translation work before the ideas are production-ready.
  • Technical depth scores 100/100, so a quick skim should focus on architecture, data, and evaluation sections before full adoption work.

Caveat

Evidence appears benchmark-centric, so verify transfer to production workloads before acting on the claims.

Estimated Reading Priority

High - 99/100 signal; read before acting on adjacent agent, evaluation, inference, or ML systems work.

Links

N/AJSON
99/100Read

VideoChat3: Fully Open Video MLLM for Efficient and Generalist Video Understanding

Published 2026-07-16 · Fetched 2026-07-17

Innovation Summary

VideoChat3: Fully Open Video MLLM for Efficient and Generalist Video Understanding: For effectiveness, we develop a scalable video data synthesis pipeline that curates three diverse, high-quality training datasets: VideoChat3-Academic2M, VideoChat3-LV116K, and VideoChat3-OL617K, covering general, long-form, and streaming.

Executive Summary

VideoChat3: Fully Open Video MLLM for Efficient and Generalist Video Understanding: For effectiveness, we develop a scalable video data synthesis pipeline that curates three diverse, high-quality training datasets: VideoChat3-Academic2M, VideoChat3-LV116K, and VideoChat3-OL617K, covering general, long-form, and streaming. Why it matters: Overall signal 99/100 driven by novelty 100 and practical impact 100. It maps to cross-cutting AI systems work even without explicit category metadata. Community signal includes 89 upvote(s) and 1 comment(s), which helps separate durable interest from title-only curiosity. Implementation angle: Implementation potential scores 99/100; prioritize adaptation paths for internal agent, evaluation, or platform workflows. No linked repository is present, so expect more translation work before the ideas are production-ready. Technical depth scores 100/100, so a quick skim should focus on architecture, data, and evaluation sections before full adoption work. Caveat: Evidence appears benchmark-centric, so verify transfer to production workloads before acting on the claims.

Why It Matters

  • Overall signal 99/100 driven by novelty 100 and practical impact 100.
  • It maps to cross-cutting AI systems work even without explicit category metadata.
  • Community signal includes 89 upvote(s) and 1 comment(s), which helps separate durable interest from title-only curiosity.

Implementation Angle

  • Implementation potential scores 99/100; prioritize adaptation paths for internal agent, evaluation, or platform workflows.
  • No linked repository is present, so expect more translation work before the ideas are production-ready.
  • Technical depth scores 100/100, so a quick skim should focus on architecture, data, and evaluation sections before full adoption work.

Caveat

Evidence appears benchmark-centric, so verify transfer to production workloads before acting on the claims.

Estimated Reading Priority

High - 99/100 signal; read before acting on adjacent agent, evaluation, inference, or ML systems work.

Links

N/AJSON
98/100Read

UniVR: Thinking in Visual Space for Unified Visual Reasoning

Published 2026-07-14 · Fetched 2026-07-17

Innovation Summary

UniVR: Thinking in Visual Space for Unified Visual Reasoning: We introduce UniVR, the first investigation into simultaneously learning complex reasoning, fine-grained physical dynamics, and long-term planning from pure visual demonstrations.

Executive Summary

UniVR: Thinking in Visual Space for Unified Visual Reasoning: We introduce UniVR, the first investigation into simultaneously learning complex reasoning, fine-grained physical dynamics, and long-term planning from pure visual demonstrations. Why it matters: Overall signal 98/100 driven by novelty 100 and practical impact 100. It maps to cross-cutting AI systems work even without explicit category metadata. Community signal includes 19 upvote(s) and 1 comment(s), which helps separate durable interest from title-only curiosity. Implementation angle: Implementation potential scores 89/100; prioritize adaptation paths for internal agent, evaluation, or platform workflows. No linked repository is present, so expect more translation work before the ideas are production-ready. Technical depth scores 100/100, so a quick skim should focus on architecture, data, and evaluation sections before full adoption work. Caveat: Evidence appears benchmark-centric, so verify transfer to production workloads before acting on the claims.

Why It Matters

  • Overall signal 98/100 driven by novelty 100 and practical impact 100.
  • It maps to cross-cutting AI systems work even without explicit category metadata.
  • Community signal includes 19 upvote(s) and 1 comment(s), which helps separate durable interest from title-only curiosity.

Implementation Angle

  • Implementation potential scores 89/100; prioritize adaptation paths for internal agent, evaluation, or platform workflows.
  • No linked repository is present, so expect more translation work before the ideas are production-ready.
  • Technical depth scores 100/100, so a quick skim should focus on architecture, data, and evaluation sections before full adoption work.

Caveat

Evidence appears benchmark-centric, so verify transfer to production workloads before acting on the claims.

Estimated Reading Priority

High - 98/100 signal; read before acting on adjacent agent, evaluation, inference, or ML systems work.

Links

N/AJSON
97/100Read

SearchOS-V1: Towards Robust Open-Domain Information-Seeking Agent Collaboration

Published 2026-07-16 · Fetched 2026-07-17

Innovation Summary

SearchOS-V1: Towards Robust Open-Domain Information-Seeking Agent Collaboration: We introduce SearchOS, a system-level multi-agent framework that turns fragile, implicit search progress into explicit, persistent, and shared state.

Executive Summary

SearchOS-V1: Towards Robust Open-Domain Information-Seeking Agent Collaboration: We introduce SearchOS, a system-level multi-agent framework that turns fragile, implicit search progress into explicit, persistent, and shared state. Why it matters: Overall signal 97/100 driven by novelty 100 and practical impact 100. It maps to cross-cutting AI systems work even without explicit category metadata. Community signal includes 44 upvote(s) and 0 comment(s), which helps separate durable interest from title-only curiosity. Implementation angle: Implementation potential scores 83/100; prioritize adaptation paths for internal agent, evaluation, or platform workflows. No linked repository is present, so expect more translation work before the ideas are production-ready. Technical depth scores 100/100, so a quick skim should focus on architecture, data, and evaluation sections before full adoption work. Caveat: No linked implementation is available yet, which raises integration cost and lowers reproducibility confidence.

Why It Matters

  • Overall signal 97/100 driven by novelty 100 and practical impact 100.
  • It maps to cross-cutting AI systems work even without explicit category metadata.
  • Community signal includes 44 upvote(s) and 0 comment(s), which helps separate durable interest from title-only curiosity.

Implementation Angle

  • Implementation potential scores 83/100; prioritize adaptation paths for internal agent, evaluation, or platform workflows.
  • No linked repository is present, so expect more translation work before the ideas are production-ready.
  • Technical depth scores 100/100, so a quick skim should focus on architecture, data, and evaluation sections before full adoption work.

Caveat

No linked implementation is available yet, which raises integration cost and lowers reproducibility confidence.

Estimated Reading Priority

High - 97/100 signal; read before acting on adjacent agent, evaluation, inference, or ML systems work.

Links

N/AJSON

Additional Papers

RoboTTT: Context Scaling for Robot Policies

Published 2026-07-16 · Fetched 2026-07-17

RoboTTT: Context Scaling for Robot Policies: We introduce Test-Time-Training Robot Policies (RoboTTT), a robot model and training recipe that scale visuomotor context to 8K timesteps, three orders of magnitude beyond state-of-the-art policies,.

95/100Read

GRASP: GRanularity-Aware Search Policy for Agentic RAG

Published 2026-07-11 · Fetched 2026-07-17

GRASP: GRanularity-Aware Search Policy for Agentic RAG: In this paper, we introduce GRASP, a reinforcement learning (RL) framework for training agents to adaptively coordinate complementary retrieval tools during multi-step reasoning.

93/100Read

From Pixels to States: Rethinking Interactive World Models as Game Engines

Published 2026-07-15 · Fetched 2026-07-17

From Pixels to States: Rethinking Interactive World Models as Game Engines: Complementing this analysis, we present a scalable data engine for Black Myth: Wukong that collects over 90 hours of gameplay with frame-aligned player actions, ground-truth game.

92/100Read

BadWAM: When World-Action Models Dream Right but Act Wrong

Published 2026-07-16 · Fetched 2026-07-17

BadWAM: When World-Action Models Dream Right but Act Wrong: We introduce BadWAM, a unified framework for modeling and evaluating World-Action Drift Attacks: a new class of WAM-specific adversarial attacks that use small visual perturbations to.

90/100Read

WanSong v1.0 Technical Report

Published 2026-07-16 · Fetched 2026-07-17

WanSong v10 Technical Report: To address these needs, we present WanSong, a simple yet powerful approach for long-form, commercial-grade song generation.

79/100Worth Watching

DeepLoop: Depth Scaling for Looped Transformers

Published 2026-07-15 · Fetched 2026-07-17

DeepLoop: Depth Scaling for Looped Transformers: The resulting method, DeepLoop, keeps the Post-LN DeepNorm architecture and sets α=(2N)^{1/2} and β=(8N)^{-1/2} for unrolled depth N.

76/100Worth Watching

Video = World + Event Stream

Published 2026-07-16 · Fetched 2026-07-17

Video = World + Event Stream: We present Wan-Streamer v0.

74/100Worth Watching

LongStraw: Long-Context RL Beyond 2M Tokens under a Fixed GPU Budget

Published 2026-07-16 · Fetched 2026-07-17

LongStraw: Long-Context RL Beyond 2M Tokens under a Fixed GPU Budget: A growing gap separates inference context lengths from RL post-training: inference systems are approaching million-token contexts, while post-training workloads often remain at 256K tokens or below.

72/100Worth Watching

Watchlist

Archive

Daily record count: 22. Persistent paper JSON lives under public data.

  1. SEED: Self-Evolving On-Policy Distillation for Agentic Reinforcement LearningPublished 2026-07-16 · 99/100 · Read
  2. Spectral Rewiring for Exploration, Purification, and Model MergingPublished 2026-07-03 · 99/100 · Read
  3. VideoChat3: Fully Open Video MLLM for Efficient and Generalist Video UnderstandingPublished 2026-07-16 · 99/100 · Read
  4. UniVR: Thinking in Visual Space for Unified Visual ReasoningPublished 2026-07-14 · 98/100 · Read
  5. SearchOS-V1: Towards Robust Open-Domain Information-Seeking Agent CollaborationPublished 2026-07-16 · 97/100 · Read
  6. RoboTTT: Context Scaling for Robot PoliciesPublished 2026-07-16 · 95/100 · Read
  7. GRASP: GRanularity-Aware Search Policy for Agentic RAGPublished 2026-07-11 · 93/100 · Read
  8. From Pixels to States: Rethinking Interactive World Models as Game EnginesPublished 2026-07-15 · 92/100 · Read
  9. Smarter and Cheaper at Once: Byte-Exact KV-Cache Grafting Turns a Frozen Small Model into a Verified-Knowledge FlywheelPublished 2026-07-15 · 91/100 · Read
  10. VIABench: A Comprehensive Video Benchmark Collected from Blind Individuals for Visual Impairment AssistancePublished 2026-07-16 · 91/100 · Read
  11. BadWAM: When World-Action Models Dream Right but Act WrongPublished 2026-07-16 · 90/100 · Read
  12. MultiRef-Compass: Towards Comprehensive Evaluation of Multi-Reference-to-Audio-Video GenerationPublished 2026-07-15 · 89/100 · Read
  13. MeanFlowNFT: Bringing Forward-Process RL to Average-Velocity GeneratorsPublished 2026-07-16 · 87/100 · Read
  14. Concurrent Image Understanding and Generation: Self-Correcting Coupled Markov Jump ProcessesPublished 2026-07-14 · 86/100 · Read
  15. Demystifying On-Policy Distillation: Roles, Pathologies, and RegulationsPublished 2026-07-15 · 82/100 · Read
  16. WanSong v1.0 Technical ReportPublished 2026-07-16 · 79/100 · Worth Watching
  17. DeepLoop: Depth Scaling for Looped TransformersPublished 2026-07-15 · 76/100 · Worth Watching
  18. Video = World + Event StreamPublished 2026-07-16 · 74/100 · Worth Watching
  19. LongStraw: Long-Context RL Beyond 2M Tokens under a Fixed GPU BudgetPublished 2026-07-16 · 72/100 · Worth Watching
  20. KeyFrame-Compass: Towards Comprehensive Evaluation of Keyframe-Conditioned Video GenerationPublished 2026-07-15 · 61/100 · Worth Watching
  21. AsySplat: Efficient Asymmetric 3D Gaussian Splatting for Long-Sequence Scene ModelingPublished 2026-07-13 · 56/100 · Skip
  22. Partition, Prompt, Aggregate: Statistical Self-Consistency in Language ModelsPublished 2026-07-16 · 54/100 · Skip